CHAPTER 198

Design wave height related to structure lifetime

Zhou Liu1 Hans F. Burcharth 2

Abstract
The determination of the design wave height (often given as the significant wave height)
is usually based on statistical analysis of long-term extreme wave height measurement or
hindcast. The result of such extreme wave height analysis is often given as the design wave
height corresponding to a chosen return period. Sometimes confidence band of the design
wave height is also given in order to include various sources of uncertainties.
In this paper the First Order Reliability Method (FORM) is used to determine the design
wave height corresponding to a certain exceedence probability within the structure lifetime.
This includes the statistical vagrancy of nature, sample variability and the uncertainty due
to measurement or hindcast error. Moreover, based on the discussion on the statistical
vagrancy of nature, a formula for the calculation of encounter probability is presented.

1 Introduction
The determination of the design wave height (often given as the significant
wave height) is usually based on statistical analysis of long-term extreme wave
height measurement or hindcast. The sources of uncertainty contributing to
the uncertainty of the design wave height are (Burcharth 1992):
1) Statistical vagrancy of nature, i.e. the extreme wave height X is a
random variable.
2) Sample variability due to limited sample size.
3) Error related to measurement, visual observation or hindcast.
4) Choice of distribution as a representative of the unknown true long-
term distribution
5) Variability of algorithms (choice of threshold, fitting method etc.)
6) Climatological changes

The sources 1, 2 and 3 and their influence on the design wave height will be
discussed in this paper.
1
Ph.D., Department of Civil Engineering, Aalborg University, Denmark
2
Professor, dr.techn., Department of Civil Engineering, Aalborg University, Denmark

2560
 DESIGN WAVE HEIGHT 2561

We use an example to demonstrate how the design wave height is determined.

The data consist of 17 most severe storms in a period of 20 years for a deep
water location. The Gumbel distribution curve in Fig.l is obtained by fitting
these 17 wave data to a Gumbel distribution.

Return value XT (m)

15
12 2
- - JP°*la \
Gumbel

100 1000
Return period T (years)

Fig.l. Design wave height.

If the design level for the design wave height is a return period of 100 years,
i.e. T — 100, the design wave height is a;100 = 12.2 m, which means that
on average this 12.2 m design wave height will be exceeded once in every 100
years.
The design wave height can be better described by the use of encounter proba-
bility, i.e. the probability that the design wave height will be exceeded within
the structure lifetime. For example if the structure lifetime L is 25 years, the
encounter probability of the design wave height xwo is

p = 1 - exp(--) = 22% (1)

Eq (1) is derived as eq (11) in section 2.

This means that the 12.2 m design wave height will be exceeded with 22%
probability within a structure lifetime of 25 years.
If the sample variability is included, the design wave height a;100 becomes a
random variable. The distribution of the design wave height a;100, which is
usually assumed to follow the normal distribution, can be obtained by numer-
ical simulation, cf. Fig.l. If the upper bound with 90% confidence is taken
as the design level, the design wave height is 14.8 m. What is the exceedence
probability of the 14.8 m design wave height within the structure lifetime
? It cannot be calculated straight away but it might be guessed that it is
2562 COASTAL ENGINEERING 1996

p = 1 - 0.9 ( 1 - 0.22 ) = 30%

In this paper the First Order Reliability Method (FORM) is used to determine
a design wave height corresponding to a certain exceedence probability within
a specified structure lifetime. This includes the statistical vagrancy of nature,
sample variability and the uncertainty due to measurement/hindcast error.
Moreover, based on the discussion on the statistical vagrancy of nature, a
formula for the calculation of encounter probability is presented.
2 Design wave height related to the statistical vagrancy of nature:
Encounter probability
Even if we had an infinite quantity of historic true wave data and knew the
related distribution precisely, there would still be uncertainty as to the largest
wave which will occur during any period of time - simply due to the statistical
vagrancy of nature. In this case the design wave height related to structure
lifetime is characterized by the encounter probability, i.e. the probability that
the design wave height will be exceeded within the structure lifetime.
Assume that the number of the extreme events is N within the structure
lifetime L. X1 denotes the maximum value in these N independent trials.
Then the distribution function of X1 is

Fxi(x) = P(x1<x) = (Fx(x)f (2)

Note that Fxi can be interpreted as the non-occurrence of the event (X > x)
in any of N independent trials.
Assuming that the number of the extreme events N = XL, where the sample
intensity A is
_ number of extreme events
number of years of observation
From the definition of the return period T

A ( 1 - Fx(x) ) ^ '
we get from eq (2)
(1 \ ^
1 - W) (5)

The encounter probability of x, i.e. the probability that x will be exceeded

within the structure lifetime L, is

p = 1 - Fx>(x) = 1 - (l - ^)AL (6)

 DESIGN WAVE HEIGHT 2563

For the case where A = 1, eq (6) becomes

(7)
However, the number of the extreme events within the structure lifetime N is
also a random variable. N is usually assumed to follow the Poisson distribution
(XL)-
P(N = n) exp(-XL) n = 0, 1, 2, (8)

The probability of the event (X1 < x or X < x) within the structure lifetime

FXi(x) = P{X1<x) = Y,lP(N = n)Fxl(x,n)

(AL)»
exp(-XL) ( Fx{x) )n
n=0 . n\

(A L Fx(a
= E exp(-XL)

(X L Fx(x) )n
= exp(-XL) Y,
n\

= exp(—X L) exp(X L Fx(x)

= exp[XL(Fx(x) - 1)] (9)

Inserting eq (4) into eq (9) is obtained

Fxi(x) = exp[-- (10)

The encounter probability of x within the structure lifetime is

p = 1 - FXi(x) = 1 exp (11)

'T,
Eq (11) is not only simpler than eq (6), but has stronger theoretical background
as well because it treats iVasa random variable.

3 Design wave height related to the statistical vagrancy of nature,

sample variability and measurement/hindcast error
To exemplify the discussion, it is assumed that the extreme wave height follows
the Gumbel distribution

F = Fx(x) = P(X<x) = exp (-exp (-(^-)j} (12)

2564 COASTAL ENGINEERING 1996

where X is the extreme wave height which is a random variable, x a realization

of X, A and B the distribution parameters.
Due to the sample variability and measurement/hindcast error, the distribu-
tion parameter A and B become random variables, and the maximum wave
height within the structure lifetime, X1, becomes a conditional random vari-
able X1\AB. The probability of X1 > x0 within the structure lifetime is

PiX^g^Xo) = P(x0-X %B<V (13)

Now consider the failure function
< 0 failure
g(x\a,b) = Xo-X^^g < = 0 limit state (14)
> 0 no failure

It can be seen that the failure probability of the failure function is actually
the exceedence probability of the design wave height xo within the structure
lifetime.
By the use of the Rosenblatt transformation, the Hasofer and Lind reliability
index (3 for the failure function can be estimated by the First Order Reliability
Theory (FORM). The failure probability, i.e. the probability of X1 > x0 within
the structure lifetime, is calculated by

P(X\B > x0) « $(-/?) (15)

where $ is the standard normal distribution. The procedure for the calculation
of /? is detailed in the Appendix.

4 Numerical simulation of aA and as

The only unknown in the calculation of /3 is the distribution of A and B.
Due to the sample variability, i.e. the influence of limited number of data, the
distribution parameters A and B, estimated from a sample, are subject to an
uncertainty.
Wave data set contains measurement/hindcast error. Measurement error is
from malfunction and non-linearity of instruments, such as accelerometer and
pressure cell, while hindcast error occurs when the sea-level atmospheric pres-
sure fields are converted to wind data and further to wave data. The accuracy
of such conversion depends on the quality of the pressure data and on the
technique which is used to synthesize the data into the continues wave field.
Burcharth (1986) gives an overview on the variational coefficient C (standard
deviation over mean value) of measurement/hindcast error.
In order to account for the sample variability and measurement/hindcast error,
A and B are assumed to follow the normal distribution. The mean values ^A
 DESIGN WAVE HEIGHT 2565

and HB are obtained by fitting the data to the distribution by one of the fitting
methods, such as maximum likelihood method or the least square method. The
standard deviations a A and <JB are obtained by numerical simulations, taking
into account the sample variability and the hindcast error, as explained as
follows:
A sample with size N is fitted to the Gumbel distribution

Fx(x) = P (X < x) = exp {-exp (-(L^))) (16)

The obtained distribution parameters Atrue and J5,rue are assumed to be the true
values. Numerical simulation is applied to get the standard deviations of the
estimators A and J5, taking into account the sample variability corresponding
to the sample size N. The procedure is as follows:

1) Generate randomly a number between 0 and 1. Let the non-

exceedence probability F equal the number, the single extreme
data x is obtained by

x = F?(F) = ArUe[-ln(-lnF)] + Btm (17)

2) Repeat step 1) N times. Thus we obtain a sample belonging to the

distribution of eq (16) and the sample size is N.
3) Fit the sample to the Gumbel distribution and get the new esti-
mated distribution parameters A and B.
4) Repeat steps 2) and 3), say, 10,000 times. Thus we get 10,000
values of A and B.
5) Calculate the standard deviations <JA and erg.

In order to include the measurement/hindcast error it is assumed that the

hindcast error follows a normal distribution. The following step can be added
after step 1).

1*) Generate randomly a number between 0 and 1. Let the non-

exceedence probability F be equal to the number. The modified
extreme data a;modifled is obtained by

zm„dified = x + C x $-\F) (18)

where $ is the standard normal distribution and C is the coefficient

of variation of the measurement/hindcast error. C ranges usually
from 0.05 to 0.2 as suggested by Burcharth (1986).
2566 COASTAL ENGINEERING 1996

5 Examples
The deep water wave data presented in Fig.l is used as an example to demon-
strate the determination of the design wave height and the influence of sample
variability and measurement/hindcast error.
The data set consists of 17 significant wave heights corresponding to the 17
most severe storms in a period of 20 years, i.e. A = 17/20. By fitting a
Gumbel distribution to the extreme data we obtain the distribution parameters
A = 1.73 and B = 4.53, cf. Fig.l.
If only the statistical vagrancy of the nature is considered, i.e. A and B are
exact values, the wave height corresponding to any return period can be found
from the graph. The 100 year return period significant wave height is 12.2 m,
which by use of eq (11) is found to correspond to 22% exceedence probability
within 25 year structure lifetime.
Sample variability
Taking into account the sample variability, the distribution parameters A and
B become random variables. Their distributions shown in Fig.2 are obtained
by the Monto-Carlo simulation as explained in section 4.

Probability density Dab .lity density

1.0- A 1.0-
Gumbel distribution h Gumbel distribution
0.8- with A= 1.73, B= 4.53 0.8- 1 with A= 1.73, B= 4.53
Sample size 17 | 1 Sample size 17
0.6- 0.6
Repeat number 15000 1 1 Repeat number 15000
0.4- 0.4
average of A 1.72 f 1 average of B 4.57
0.2- standard dev of A 0.418 0.2- \ \ standard dev of B 0.449

1 i
i • i i > i
0.0 2.4 4.8 7.2 9.6 12.0 5.4 7.8 10.2 12.6 15.0

Distribution parameter A Distribution parameter B

Fig.2. Distribution of the Gumbel parameters A and B by sample variability

(N=n).
The probability density and the non-exceedence probability of the maximum
significant wave height within any structure lifetime can be estimated by
FORM. Fig.3 shows the results for a structure lifetime of 25 years. The figure
includes for comparison also graphs where the sample variability is omitted.
These graphs are obtained by eq (11).
 DESIGN WAVE HEIGHT 2567

Non-exceedence Prob.

Statistical vagrancy
Statistical vagrancy +
sample variability

Sample size N=17 Statistical vagrancy +

sample variability

10 12 14 16 IB
Maximum significant wave height within Maximum significant wave height within
structure lifetime of 35 years structure lifetime of 25 years

Fig.3. Distribution of maximum significant wave height (sample size N=17).

If the design level is the significant wave height corresponding to 22% excee-
dence probability within 25 years (T = 100 years), it can be seen from Fig.3
that the design wave height with consideration of the sample variability is
12.7 m, which is a little larger than the value without the consideration of the
sample variability (12.2 m). It can also be seen that the design wave height
of 14.8 m (upper bound with 90% confidence, cf. Fig.l) corresponds to 9%
exceedence probability within 25 years, not 30% as guessed in Section 1.
In the case of a bigger sample size, e.g. TV = 100, there is almost no difference
between the design wave height with and without sample variability, cf. Fig.4.
For comparison the same A value is applied.

Non-exceedence Prob,
1.00

Statistical vagrancy
Statistical vagrancy +
sample variability

Sample size N=100

Maximum significant wave height within Maximum significant wave height within
structure lifetime of 25 years structure lifetime of 25 years

Fig.4- Distribution of maximum significant wave height (sample size N=1Q0).

Table 1 shows the design wave height corresponding to different sample size.
Sample size oo means that there is no sample variability. Keep in mind that a
2568 COASTAL ENGINEERING 1996

typical sample size is about 20, it can be said that sample variability has some
limited influence on the design wave height.

Table 1. Significant wave height corresponding to p = 22% within L •= 25 years

sample size 10 17 50 100 CO

Hs(m) 13.0 12.7 12.4 12.3 12.2

relative difference 6.5% 4% 1.6% 0.8% 0

Measurement/hindcast error
The same procedure can be applied to further include the measurement/hindcast
error.
The variational coefficient of the extreme data listed in Table 2 is taken from
Burcharth (1986). Data based on visual observation from ships should in
general not be used for determination of design wave height because ships
avoid poor weather on purpose. With the advances in measuring techniques
and numerical models, generally the C value has been reduced to app. 0.1 or
less.

Table 2. Coefficient of variation for significant wave height (Burcharth 1986).

Methods of determination Coefficient of variational

Accelerometer buoy
Pressure cell 0.05 - 0.1
Vertical radar

Horizontal radar 0.15

Hindcast, SPM method 0.15 - 0.2

Hindcast, numerical 0.1 - 0.2

Visual observation 0.2

In Fig.5 the coefficient of variation C of the extreme data due to measure-

 DESIGN WAVE HEIGHT 2569

ment/hindcast error is assumed a typical value of 0.1.

Non—exceedence Prob.

0.80 -

Sample size N=17

Statistical vagrancy

0.40 Statistical vagrancy +

Sample variability

0.20 - •*-++ Statistical vagrancy +

sample variability +
hindcast error ( C=0.1)

0.00 -±*
6 8 10 12 14 18 IB 20 22
Maximum significant wave height within
structure lifetime of 25 years

Fig.5. Distribution of maximum significant wave height (N = 17, C — O.lJ.

In Table 3 is given values extracted from Fig.5 corresponding to an exceedence

probability of p = 22%.

Table 3. Significant wave height corresponding to p = 22% within L = 25 years

Case H.(m) Remarks

Statistical vagrancy 12.2 m

Statistical vagrancy +
sample variability 12.71 m sample size N — 17

Statistical vagrancy +
sample variability + sample size N = 17
hindcast error 12.75 m variational Coeff. C = 0.1

It can be seen from Fig.5 and Table 3 that the influence of measurement/hindcast
error on the design wave height is very small.
2570 COASTAL ENGINEERING 1996

6 Conclusions
The paper concentrates on the exceedence probability of the design wave height
within the structure lifetime ( encounter probability ).
If only the statistical vagrancy of the nature is included, a new and simple en-
counter probability formula is derived which takes into account the randomness
of the extreme events within the structure lifetime.
If other uncertainties should be considered, the paper shows that the reli-
ability theory can be applied to determine the encounter probability of the
design wave height. A practical example shows that normally sample vari-
ability has little influence on the design wave height, while the influence of
measurement/hindcast errors is almost negligible.

7 References

Burcharth, H.F. , 1986. On the uncertainties related to the estimation of ex-

treme environmental conditions. Proceeding of Seminar on Uncertainties
Related to the Design and Construction of Offshore Jacket Structures,
Copenhagen, 1986, Published by Danish Society of Hydraulic Engineer-
ing
Burcharth, H.F. and Zhou Liu , 1994. On the extreme wave height analy-
sis. Proceedings of HYDRO-PORT'94, Yokosuka, Japan, 19-21 October,
1994
Goda, Y. , 1988. On the methodology of selecting design wave height. Proc.
21st Int. Conf. on Coastal Engr., Spain.
Thoft-Christensen, P. and Michael J. Baker , 1982. Structural reliability the-
ory and its application. ISBN 3-540-11731-8, Springer-Verlag, 1982.
 DESIGN WAVE HEIGHT 2571

Appendix: Estimation of reliability index /?

The followings explain the procedure for the calculation of /?.
From eq (9) is obtained
Fxi(x1) = exp[XL(Fx(x1) - 1)]

-B.
exp XL exp I —exp (19)

which can be rewritten as

x1 = A In -In 1 + + B (20)
XL
X1 can be converted to the standard normal distributed random variable U1
by
$(Ul) = Fx^x1) (21)
Inserting eq (21) into eq (20) is obtained
ln$(«i)
x1 = A - In I - In I 1 + + B (22)
XL
The failure function becomes
ln$(ui~
g(u\,a,b) — XQ — A —In In 1 + B (23)
XL
The normal random variables A and B are converted into the standard normal
distributed random variables U2 and Us respectively
A - fiA _ B - flB
u2 = M3 (24)

Insert eq (24) into eq (23) is obtained

ln$(wi)
g(v,i,u2,ua) = x0 - (HA + VA^) -In -In 1 +
XL

~ {HB + &B Us) (25)

The differentiations of the failure function are
dg _ (fiA +<*A u2) <t>(ui)
at
dut In ( 1 + ±±&1) ( 1 + *4£*1) A L $(Ul)

dg_ ln$(«i)
«2 -<?A In -In 1 + (26)
du2 XL

dg_
0-3 = -0B
du3
2572 COASTAL ENGINEERING 1996

where <f> is the density function of the standard normal distribution.

The iterative procedure for calculation of /3 is

1) Select trial values : u* = (uj, u\, u%).

2) Insert u* into eq (26) and get (ax, a-i, a3).
3) Determine a better estimate of u* by
3

U* = Oi ^
3

4) Repeat steps 2) and 3) to achieve convergence.

5) Calculate /3 by
i
3
2
ft = (E «) )